JP2001014349A - Rating system for collaboration information filtering - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a rating system for collaboration information filtering with a high function that a user easily can use. SOLUTION: The system has a function of reflecting the user's viewpoint of values, automatically specifying similar users from many aspects of view, and displaying the evaluation so that the user easily can see. According to a viewpoint that a request user specifies, the scoring history of the user as to an object item to be evaluated in a personal profile 6 is analyzed to calculate the similarity of the user, and according to the similarity, an item suitable for the request user is recommended. The viewpoint of the request user is specified with a viewpoint vector showing the evaluation object item that the request user regards as an important item and automatically extracted. On the basis of the similarity, users similar to the request user are automatically determined.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to the technical field of information filtering, and more particularly, to selecting information suitable for a person based on the evaluation history of another user and the evaluation history of the person from many pieces of information. And a collaborative filtering method that supports

[0002]

2. Description of the Related Art Information filtering technology is an elemental technology for selecting appropriate information from a vast amount of information. With the spread of the Internet and the like, various contents and information around the world are distributed. Nowadays, it is an important issue how to select the information that I really need from the vast amount of information.

[0003] Information filtering techniques can be broadly classified into a content-based filtering technique for analyzing and selecting information contents and a collaborative filtering technique for selecting information using evaluations of a large number of users. Most conventional filtering systems are content-based. For example, when a document to be filtered is a document, the filtering is performed by analyzing attributes and characteristics of the content to be filtered using keyword frequency information or the like. However, in content-based filtering, (1) information to be filtered is limited to information that can be understood by a computer, and (2) a content-based system can make only objective judgments and cannot make subjective judgments. There are problems such as (3) there is no discovery of something in a content-based system.

[0004] In contrast, collaborative filtering techniques are based on the idea that "the human being is the best judge for subjective characteristics such as trust and quality", and provide feedback from other users in the community. The above problem is solved by making the best use of it, and applying a collaborative filtering method such as word-of-mouth and information sources to the system. In this technique, other similar users who are interested in information of interest are specified, and information is selected based on their opinions. The identification of similar users includes those performed by the user himself and those automatically performed by the system. In the method of identifying similar users on their own, it is necessary to specify in advance other users who are likely to be interested in the information that they are interested in, and a small community where users know each other well Only works properly, and the user has considerable responsibility. On the other hand, in the method that the system automatically identifies,
The challenge is how to automatically identify similar users, and several methods have been proposed for users to rate information quality and for the system to identify similar users based on the similarity of the scoring history. However, no established method is known.

In a conventional method, a method of specifying a similar user based on a user's evaluation history is performed by calculating a least square error of the scoring history of each user based on the user's scoring history for a plurality of items (items). , A method of specifying a user whose error is less than a threshold as a similar user, or a method of calculating a correlation coefficient of a scoring history of each user and specifying a user whose correlation coefficient is equal to or more than a threshold as a similar user and so on. When the similar user is specified in this way, the system can recommend the item to the user by presenting the item that has been highly evaluated in the evaluation of the similar user to the user.

[0006]

As described above, in the collaborative filtering, when it is difficult for a user to specify another user having the same interest in advance, the system automatically specifies a similar user. And presenting their assessments may be beneficial. However, the conventional technique for identifying similar users has the following problems.

First, the calculation method for specifying similar users according to the conventional method described above does not consider the relationship between items to be scored at all. It is premature to assume that the opinions of two users agree on all types of items simply because they agree on one type of item. For example, in the case of music, even if they both like classical music,
You may have the opposite sensibility when it comes to jazz. Thus, two users may have multiple categories, in some categories they strongly agree, in some al categories they do not agree at all, and in some categories there is no opinion correlation . If the simple average of all items is calculated without paying attention to such category types, the probability that the system will make a correct evaluation will decrease.

In the conventional calculation method for specifying similar users, the system determines a set of similar users, so that only a single viewpoint can be recommended. But,
In practice, it is preferable that a plurality of sets of similar users are specified from various viewpoints and recommendations are made from various viewpoints.

Further, the calculation method for identifying similar users according to the conventional method does not reflect user values such as what kind of evaluation is regarded as important. In order to properly select similar users, simple minimum error square method or correlation coefficient method for all items is not enough, and the values of users such as which items are regarded as important are reflected. Should be.

[0010] Alternatively, the opinion of a person having a completely opposite taste may be more important than the opinion of a user similar to the user. It is possible that a new perspective can be created by touching the opinion of a completely different person, but the conventional method cannot convey to a user the opinion of a person who has a completely opposite taste to the user.

An object of the present invention is to provide a user-friendly and high-functionality function having a function of automatically identifying similar users from various viewpoints reflecting user values and presenting their evaluations to the user in an easy-to-understand manner. It is to provide a rating system for collaborative information filtering.

[0012]

In order to solve the above-mentioned problems, the present invention provides a requesting user by presenting evaluation information of a plurality of items to be evaluated by a plurality of users to a requesting user upon receiving the request from the user. A user request processing unit that processes so that an appropriate item can be selected from the plurality of evaluation target items, and, based on a viewpoint specified by the requesting user, the scoring history of the plurality of users with respect to the evaluation target item. Analyzing the similarity between the plurality of users by analyzing the plurality of users, and recommending an item considered appropriate for the requesting user based on the similarity. Provide a rating system.

Further, according to the present invention, the viewpoint of the requesting user is as follows:
The evaluation history analyzer is specified by a viewpoint vector indicating an evaluation target item considered to be important by the requesting user, and the evaluation history analysis unit specifies the evaluation value and the viewpoint vector given to each evaluation target item by each user. Calculating the similarity to be obtained.

[0014] In the present invention, the evaluation history analysis unit may include a scoring value given to each evaluation target item by each user and all scoring values given to each evaluation target item by every user. And calculating the similarity defined by the difference from the average value.

Further, in the present invention, the evaluation history analysis unit automatically determines a similar user of the requesting user based on the similarity.

Further, in the present invention, the evaluation history analysis unit recommends an item which is highly scored by a user determined as a similar user.

Further, in the present invention, the evaluation history analysis unit predicts an evaluation of an evaluation target item that is not scored by each user based on the similarity and an evaluation of the evaluation target item by another user. And recommending the item.

Further, in the present invention, the evaluation history analysis unit also determines a ranking of the evaluation target item based on a score given to the evaluation target item by a plurality of users, and the user request processing unit And presenting ranking information on the ranking determined by the evaluation history analysis unit to the requesting user.

Further, in the present invention, the user request processing unit includes, along with the ranking information, the number of users who scored each evaluation target item, the average score of each evaluation target item, the variance of the score of each evaluation target item, and The present invention is also characterized in that related information including comments and recommendation information given by a user who has graded an evaluation target item is also presented.

In the present invention, the viewpoint of the requesting user is as follows:
It is specified by a viewpoint vector indicating an evaluation target item regarded as important by the requesting user, and enables the requesting user to arrange a display object representing the evaluation target item around the origin on a two-dimensional plane. A viewpoint visualization unit that automatically extracts a viewpoint of the requesting user and automatically converts a distance between the origin and each evaluation target item into a component of the viewpoint vector.

In the present invention, the similarity calculated by the evaluation history analysis unit is converted into a distance, and a display object representing the plurality of users is displayed on a two-dimensional plane while maintaining a distance relationship between the plurality of users. It is characterized by further comprising a relationship visualization unit for displaying the similarity of the plurality of users in the form of a similarity network obtained by arranging the plurality of users.

Further, the present invention provides: (a) receiving a request from a user, presenting evaluation information of a plurality of evaluation target items by a plurality of users to the requesting user so that the requesting user can obtain the evaluation information from the plurality of evaluation target items; (B) analyzing the scoring histories of the plurality of users with respect to the evaluation target item based on a viewpoint specified by the requesting user; Calculating a similarity between the two, and recommending an item that is deemed appropriate for the requesting user based on the similarity.

In the present invention, the viewpoint of the requesting user is as follows:
It is specified by a viewpoint vector indicating an evaluation target item regarded as important by the requesting user. In the step (b), a score value given to each evaluation target item by each user and the viewpoint vector are used. The specified similarity is calculated.

Further, in the present invention, in the step (b), an average of a score value given to each evaluation target item by each user and an average of all score values given to each evaluation target item by all users. Calculating the similarity defined by a difference from the value.

In the present invention, the step (b) is characterized in that a similar user of the requesting user is automatically determined based on the similarity.

Further, in the present invention, in the step (b), an item having a high score by the user determined as the similar user is recommended.

In the present invention, in the step (b), the evaluation of the evaluation target item not scored by each user is predicted based on the similarity and the evaluation of the evaluation target item by another user. And recommending the item.

In the present invention, in the step (b), the ranking of the evaluation target item is also determined based on a score given to the evaluation target item by a plurality of users, and in the step (a), , The step (b)
Is presented to the requesting user with ranking information related to the ranking determined in.

In the present invention, in the step (a), together with the ranking information, the number of users who scored each evaluation target item, the average score of each evaluation target item, the variance of the score of each evaluation target item, and The present invention is also characterized in that related information including comments and recommendation information given by a user who has graded each evaluation target item is also presented.

In the present invention, the viewpoint of the requesting user is as follows:
It is specified by a viewpoint vector indicating the evaluation target item regarded as important by the requesting user, and by enabling the requesting user to place the display object of the evaluation target item around the origin on a two-dimensional plane. The method further comprises automatically extracting a viewpoint of the requesting user, and automatically converting a distance between the origin and each evaluation target item into a component of the viewpoint vector.

Further, in the present invention, the similarity calculated in the step (b) is converted into a distance, and the display objects of the plurality of users are arranged on a two-dimensional plane while maintaining the distance relation of the plurality of users. And displaying a similarity relationship of the plurality of users in the form of a similarity network obtained by performing the operation.

[0032]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the rating for collaborative information filtering according to the present invention will be described below with reference to FIGS.

FIG. 1 shows a schematic configuration of a rating system according to this embodiment. The rating system 1 includes a user request processing unit 2, an evaluation history analysis unit 3, a relationship visualization unit 4, a content directory 5, a personal profile 6, and a rating table 7.

In the configuration shown in FIG.
Is based on http (hypertext transfer protocol) requests from users (content rating, registration, ranking information acquisition, search for similar user groups,
(Requests such as search) are passed to the evaluation history analysis unit 3, and the processing result of the evaluation history analysis unit 3 is converted to HTML (HyperText Markup
guage) This is a processing unit that plays the role of a gateway that is converted to the format and sent back to the user.

The evaluation history analysis unit 3 includes the user request processing unit 2
, The rating information of each user is totaled, statistical analysis is performed, (1) presentation of rating information of each content based on voting results, and (2) similar interests or values. Introduction of user groups that have
(3) Items (items) deemed appropriate for the user
Is a processing unit that makes a recommendation.

The personal profile 6 is a file storing a user name and a password of each user, history information of the rating so far, a keyword list indicating personal interest and an object of interest, and the like.

The rating table 7 is a table in which rating information of each content by the user is totaled. The rating table 7 is updated each time a user rates content. Further, based on the rating table 7, average evaluation of each content, introduction of people who perform evaluations similar to the user, recommendation of content deemed appropriate for the user, and the like are performed.

The contents directory 5 is a directory for storing information on actual contents to be rated.

The relation visualization unit 4 automatically extracts the user's interests and values from the operation performed by the user, determines from what viewpoint the data in the rating table 7 is to be analyzed, This is a processing unit that provides a user interface for visually displaying an analysis result such as the above.

Hereinafter, the configurations and operations of the user request processing unit 2, the rating table 7, the evaluation history analysis unit 3, and the relationship visualization unit 4 will be described in detail.

<User Request Processing Unit 2> FIG. 2 shows an example of a processing request (communication protocol with the evaluation history analysis unit 3) handled by the user request processing unit 2. Through the user request processing unit 2, the user can make requests such as user registration, scoring of various contents, obtaining ranking information, searching for and searching for similar user groups, and the like.

FIG. 3 shows an example of a user registration screen in the user request processing section 2. The user can register his / her user name, password, his / her field of interest, and the like on this user registration screen. Registration information input through this user registration screen is stored in the personal profile 6.

FIG. 4 shows an example of a scoring form used for scoring contents. The user can score the quality of each content on a five-point scale, for example, based on his subjective judgment. Based on this scoring information, the evaluation history analysis unit 3 updates the rating table 7 and can perform processing as needed.

When the user makes a ranking information acquisition request to the user request processing unit 2, the request is passed to the evaluation history analysis unit 3, where the ranking is calculated based on the evaluation histories of all users, and the calculation result is obtained. Is returned to the user request processing unit 2. Based on the calculation result, the user request processing unit 2 displays information such as the total score of the evaluation, the average score, the number of graders, and the comments of the graders to the user in HTML format.

When the user makes a similar user group search request to the user request processing unit 2, this request is passed to the evaluation history analysis unit 3, and when a similar user matching the request is specified, the specified similar user Are visually displayed through the relation visualization unit 4.

<Rating Table 7> FIG. 5 shows an example of the rating table 7. The rating table 7 is a table in which the scoring information of each evaluation item (for example, content) by each user is totaled. Each time each user scores an evaluation item, the rating table 7 is updated. Also, this rating table 7
Based on the above, the evaluation history analysis unit 3 performs an average evaluation of each evaluation item, introduces people who perform evaluations similar to the user, recommends evaluation items that are deemed appropriate for the user, and the like. The rating is, for example, a five-point scale (1 = poo
r, 3 = normal, 5 = excellent).

Various scoring criteria can be considered depending on the scoring target. As a typical content rating on the Internet, RSACi (Re
creational Software Advisory Council on the Intern
et: Recreational Software Advisory Council Internet Rating Standards: http://rsac.org
and so on. RSACi has categories of violence, nudity, sex, and language, and content on the Internet can be rated with a rating value of 0 to 4 for each of these categories. RS
ACi is a rating standard by a non-profit organization, so it has become a substantial standard for self-rating in the United States, and international rating standards are being studied based on this standard. Also, W3
C (World Wide Web Consortium) is a technical standard PICS (Platform for Internet Cont.) To standardize the format of rating information and protocol.
ent Selection), and proposes a framework for making rating information developed with various rating standards available to various filtering software.

For example, the evaluation as to whether or not the contents are harmful may differ depending on the social background, but it is unlikely that extreme differences will occur depending on the person.
In such a case, the average evaluation of many users seems to have some meaning. However, in the evaluation of digital contents such as music and videos, individual subjectivity and values are considered to occupy a large proportion. What was the worst rating for one person might be the highest rating for another. In such a case, the average of the evaluations of all members does not make much sense, but rather, the actual evaluations of those who have a tendency to evaluate with oneself are likely to be helpful. As described above, it is important to adopt an appropriate scoring method and an evaluation method of the tally result according to the target of scoring or evaluation.

<Evaluation History Analysis Unit 3> FIG. 6 shows a schematic configuration of a basic processing mechanism of the evaluation history analysis unit 3. As shown in FIG. 6, the evaluation history analysis unit 3 includes a ranking calculation unit 3
1. User similarity calculation unit 32 and recommendation algorithm 3
3 Hereinafter, the operation of each unit will be described in detail.

<Ranking Calculation Unit 31> The ranking calculation unit 31 totals the information of the rating table 7 and ranks the evaluation items. The ranking is performed in consideration of the following evaluation indices.

(1) Number of votes: This is a number indicating how many people scored each evaluation item. If this number is large for a certain evaluation item, the more people have scored, the more the topicality (attention) of this evaluation item can be considered. However, if the average score is low even if the number of votes is large, it means that attention was paid but the overall evaluation was not very good. Also, if the number of votes is small, even if the average score is good, the reliability of the number is considered to be low. For example, in the example of FIG. 5, the number of votes of Item3 is 3, and the evaluation is made by the most people.

(2) Average score: The average score of all the people who scored each evaluation item. If the average rating is high but the number of votes is low, the number is considered unreliable. On the other hand, even if the average score is low, the evaluation in a certain group may be extremely high. For example, in the example of FIG.
Has an average score of 5.0, which is the highest score.

(3) Variance: A value indicating the degree of variation in the evaluation of the persons who scored each evaluation item. It shows how much rating varies by person, and even if the average score is low, if the variance is large, it may be very valuable for a group. For example, in the example of FIG. 5, while the variance of Item1 is 0,
The variance of Item2 is 2.0. Item1 is User1 and Us
er2 both scored 5 points, while Item2
In the case of (1), User2 scored 5 points and User3 scored 1 point.

(4) Comments and recommendation information: The comments and recommendation information of the people who scored each evaluation item are useful for transmitting evaluations that cannot be expressed only with numbers to others. Even if the average evaluation is not necessarily high, if there is a strong recommendation or an appropriate comment from a person with the same values as the user, it is highly likely that the evaluation item is useful information for the user. Therefore, not only ranking information,
Evaluator comments and recommendation information are also displayed.

(5) Overall score: This is when the scoring range is 1
In the case of 5 points from the point, the calculation is performed using a method of adding (the score of each individual-2.5) to the score. In this case, if the person's score is 3 or more, it works as a positive vote, and if it is less than 3, it works as a negative vote. For example, in the example of FIG. 5, the total score of Item1 is 5
In the first place, the second place is Item5 (2.5 points), the third place
Item 3 is Item 3 and Item 4 (1.5 points), and Item 4 is Item 2 and It
em6 (1.0 point). In this way, when ranking is displayed, the highest score is displayed in this score calculation, but at the same time, the number of votes, average score, variance, commenter's comments and recommendation information are also displayed. Regarding the ranking method, as described above, information evaluated from a diversified viewpoint is provided at the same time, instead of the sole evaluation standard.

<Inter-User Similarity Calculation Unit 32> The inter-user similarity calculation unit 32 calculates the similarity between users based on the scoring history of each user. Examining the similarity between users based on the rating history performed by each user is useful in the following three points.

(1) By referring to the judgment of a person whose values are similar to those of himself / herself, there is a possibility that information that is more accurate and more accurate for himself / herself than the judgment based on the average evaluation of all members may be filtered.

(2) There is a possibility that a new perspective can be found (by knowing the existence of various values) by looking at the evaluation of another person based on a different criterion from that of himself / herself.

(3) By discovering people with similar interests or interests, there is a possibility that community formation and meeting with others can be supported.

Hereinafter, a method of calculating the similarity will be described using an example.

FIG. 7 shows a rating value of 1 to 5 (1:
Here is an example that simply shows the opinions of three users on seven items using No interest at all: 5: Very interested. The calculation of the similarity between users will be described using this example.

First, the user is required to input a viewpoint vector W = (w1, w2, w3, w4, w5, w6, w7) indicating which of the evaluation items from Item 1 to Item 7 is regarded as important. . Here, each element from w1 to w7 takes a real value from 0 to 1. Actually, when inputting each element of the viewpoint vector from the user, the user does not need to input a numerical value directly.
When the items are arranged in the dimensional space, the viewpoint vector is automatically generated by reading the item arrangement information by the system. For example, if the user considers that all Items are equally important, the viewpoint vector W = (1,1,1,1,1,1,1)
1). Conversely, when the user pays attention only to, for example, Item4, the viewpoint vector W = (0,0,0,1,0,
0,0). Also, for example, if the user sets Item1 to Item1
, Item3, Item5 and Category7, and I
In the case of classifying into two categories, ie, category 2 consisting of tem2, Item4 and Item6, and paying attention to only category 1, the viewpoint vector W = (1,0,1,0,1,0,1)
Becomes The calculation of the similarity between users is based on the viewpoint vector W
It is performed based on

As described above, by introducing a viewpoint vector indicating which element the user focuses on, and having the user directly input, it becomes possible to find similar users from multiple viewpoints.

The similarity between users can be calculated by determining the other users who are evaluating the same item as the user (when the user is User1, User2 and
User3) is defined by the following equation (1).

[0065]

[Number 1] R (x, y) = Σ i {Wx (i) Wy (i) (Ux (i) -3) (Uy (i) -3)} / (Σ i {Wx (i) 2 ( Ux (i) −3) ² } × Σ _i {Wy (i) ² (Uy (i) −3) ² }) ^1/2 (1)

Here, R (x, y) is the similarity between User-x and User-y, Ux (i) is the score of User-x for Item-i, and Wx (i) is , If User-x scores Item-i, W
_i is a coefficient representing _i (the i-th component of the viewpoint vector W described above), otherwise, is a coefficient representing 0, and Σ _i represents a sum of all the index values i of the item. still,
The value “3” appearing in the equation (1) is an average value of 1 and 5. Here, −1 ≦ R (x, y) ≦ 1, and the closer this value is to 1, the greater the positive correlation between x and y is (similar).
Conversely, the closer this value is to -1, the larger the negative correlation between x and y is (it is not similar).

The above equation (1) is the point where the viewpoint vector W is introduced and the average value of 1 and 5 from each scored value. ”
This is different from the general formula for calculating the correlation coefficient in that 3 ”is subtracted. In the general calculation method for the correlation coefficient, instead of“ 3 ”, the average of each user's evaluation is subtracted. In such a general method, for example, a person who gives a low evaluation to two items and a person who gives a high evaluation to these two items may have a similarity of 1.
This is obviously unnatural, but the above equation (1) can solve this problem.

Taking the case of W = (1,1,1,1,1,1,1,1) as an example, calculating the similarity between users using the above equation (1) gives the following.

R (User1, User2) = 0 R (User1, User3) = 1 R (User2, User3) = 0 That is, in this case, User1 and User3 are very similar to each other.

Taking the case of W = (0, 1, 0, 1, 0, 1, 0) as an example, calculating the similarity between users using the above equation (1) gives the following.

R (User1, User2) = 1 R (User1, User3) = 1 R (User2, User3) = 0 That is, in this case, User1 and User2 are very similar to each other, and User1 and User3 are also very similar. You are doing it.

As described above, by performing the similarity calculation using the viewpoint vector W, it becomes possible to specify similar users from various viewpoints.

<Recommendation Algorithm 33> The recommendation algorithm 33 predicts the evaluation of each unrated column based on the following steps, and recommends an item having a high predicted evaluation value to the user.

(33-1) Generation of similarity matrix:
Based on the similarity calculation by the user similarity calculation unit 32, a similarity matrix between all users is generated. When applied to the case where W = (0,1,0,1,0,1,0), a similarity matrix as shown in FIG. 8 is obtained.

(33-2) Determination of Similar Set: The above (3)
Using 3-1), a similar set (Neighb
orhood: N {User-i}). That is, for each user, others whose similarity is equal to or greater than the threshold value L are included in the list of similar sets. In this example, if L = 0.5, the similar set is as follows.

N ｛User1｝ = ｛User2, User3｝ N ｛User2｝ = ｛User1｝ N ｛User3｝ = ｛User1｝ (33-3) Estimation of evaluation (33-2), (33-3) To estimate the rating of each user in the unrated column. This is performed as follows.

[0077] (33-31) First, according to the following equation (2), calculates the weight W _ik for user-i for each user-k in similar set.

[0078]

W _ik = (R _ik −L) / (1−L) (2)

Where L is a threshold, _Rik is user-i and us
Represents the similarity with er-k. Here, 0 ≦ W _ik ≦ 1, and
W _ik is 0 when R _ik = L and 1 when R _ik = 1.

[0080] (33-32) according to the following equation (3), the average of the weighted evaluation of all similar users and calculations for Item-j for user-i, the evaluation expected P _ij.

[0081]

[Number 3] _{P ij = Σ k W ik U} kj / Σ k W ik (3)

Here, U _kj represents a score value for Item-j of user-k.

When a predictable evaluation score is predicted based on this algorithm, a table as shown in FIG. 9 is obtained. Here, the value enclosed in parentheses <> is an expected evaluation score.

(33-4) Recommendation: Based on the table in FIG.
The system recommends an item having a high expected score to the user, for example, as follows.

Recommendation for User2: Item6 recommendation degree 5 recommender User1 (5 points) Recommendation for User3: Item2 recommendation degree 5 recommender User1 (5 points), User2 (5 points) Item6 recommendation degree 5 recommender User1 (5 points) , User2 (5 points) <Relationship Visualization Unit 4> FIG. As shown in FIG. 10, the relationship visualization unit 4 includes a user viewpoint extraction module 101 and a similar user network display module 102.

The user viewpoint extraction module 101 is a module for extracting the above-described viewpoint vector W from the user. As shown in FIG. 11, the user freely arranges the Item boxes within a concentric circle having a radius of 1 on a two-dimensional plane. At this time, the user places items that are considered important near the origin. User viewpoint extraction module 1
01 converts the distance r _i between each Item box and the origin into the i-th component W _i of the viewpoint vector W. This conversion is performed, for example, based on the following equation (4).

[0087]

[Number 4] _{W i = 1-r i (} 4)

By using such an interface, the system can automatically generate a viewpoint vector W by freely arranging Items that the user considers important according to his / her own values. .

Similar user network display module 1
02 is an inter-user similarity calculation unit 3 as shown in FIG.
The similarity r between users calculated in step 2 is converted into a distance d, and a similarity network between users is displayed on a two-dimensional plane. At this time, a user having a large similarity value is arranged so as to be close to himself / herself. The conversion of the similarity r into the distance d can be performed using, for example, the following equation (5).

[0090]

D = 1−r (5)

FIG. 12 shows an example of the similarity network between users.
Shown in

In this way, the distance between the two users is calculated, and the users are arranged on a two-dimensional plane so as to keep the restriction on the distance. By presenting such a network to the user, the user can intuitively grasp the similarity between the users.

The embodiments described above can be applied and modified in various ways without departing from the novel and advantageous features of the invention, other than those already described above, and such applications and modifications are also It is within the scope of the claims.

[0094]

As described above, according to the present invention,
A function that automatically identifies similar users from multiple perspectives by reflecting user values and introduces their evaluations to users in an easy-to-understand manner has been introduced. A simple rating system can be realized.

Further, according to the present invention, it is possible to automatically calculate the similarity between users based on the similarity of the evaluation and scoring histories of the users and the viewpoint vector indicating the evaluation target to which the user pays attention. Because it is possible, it is possible to find similar users from various viewpoints according to the values of the users.

Further, since the similarity between users is expressed in a visually understandable manner as a network in a two-dimensional space according to the viewpoint of the user, the user can evaluate various others from various viewpoints. Can be shared.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration of a rating system according to an embodiment of the present invention.

FIG. 2 is a table showing an example of processing requests handled by a user request processing unit in the rating system of FIG. 1;

FIG. 3 is a view showing an example of a user registration screen used by a user request processing unit in the rating system of FIG. 1;

FIG. 4 is a diagram showing an example of a scoring form used in a user request processing unit in the rating system of FIG. 1;

FIG. 5 is a view showing an example of the contents of a rating table in the rating system of FIG. 1;

FIG. 6 is a diagram showing a configuration of an evaluation history analysis unit in the rating system of FIG. 1;

FIG. 7 is a view showing a table showing an example of scoring for explaining similarity calculation in the rating system of FIG. 1;

FIG. 8 shows a viewpoint vector W = (0, 1, 0, 1, 0, 1, 1).
FIG. 8 is a diagram showing a similarity matrix between users in FIG. 7 when using (0).

FIG. 9 shows a viewpoint vector W = (0, 1, 0, 1, 0, 1, 1).
FIG. 8 is a diagram showing a table showing expected evaluations to be added to the evaluation result example of FIG.

FIG. 10 is a diagram showing a configuration of a relationship visualization unit in the rating system of FIG. 1;

FIG. 11 is a diagram showing an example of an input interface of a viewpoint vector W used in a relation visualization unit in the rating system of FIG. 1;

FIG. 12 is a diagram showing an example of a similarity network expressed on a two-dimensional plane used in the relationship visualization unit in the rating system of FIG. 1;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Rating system 2 User request processing part 3 Evaluation history analysis part 4 Relation visualization part 5 Content directory 6 Personal profile 7 Rating table

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Takashi Kido Malaysia 63000 Selangor Dar Le Athan Cyberjaya Jalan Apec 43000 F-term (reference) 5B075 ND20 PQ02 PQ74 PR03 PR06 QM08

Claims (20)

[Claims] 1. Upon receiving a request from a user, presenting evaluation information of a plurality of evaluation target items by a plurality of users to the requesting user, the requesting user can select an appropriate item from the plurality of evaluation target items. A user request processing unit that performs processing as described above, and calculates a similarity between the plurality of users by analyzing a scoring history of the plurality of users with respect to the evaluation target item based on a viewpoint specified by the requesting user. A rating history analysis unit for recommending items considered to be appropriate for the requesting user based on the similarity. 2. The viewpoint of the requesting user is specified by a viewpoint vector indicating an evaluation target item regarded as important by the requesting user. The rating system according to claim 1, wherein the similarity defined by a given score value and the viewpoint vector is calculated. 3. The evaluation history analysis unit includes: a scoring value given to each evaluation target item by each user; and an average value of all scoring values given to each evaluation target item by any user. The rating system according to claim 1, wherein the similarity defined by a difference between the two is calculated. 4. The rating system according to claim 1, wherein the evaluation history analysis unit automatically determines a similar user of the requesting user based on the similarity. 5. The rating system according to claim 4, wherein the evaluation history analysis unit recommends an item having a high score by a user determined as a similar user. 6. The evaluation history analysis unit predicts an evaluation of an evaluation target item that has not been scored by each user based on the similarity and the evaluation of the evaluation target item by another user. The rating system according to claim 5, wherein the rating is recommended. 7. The evaluation history analysis unit also determines a ranking of the evaluation target item based on a score given to the evaluation target item by a plurality of users. The rating system according to claim 1, wherein ranking information on the ranking determined by the analysis unit is presented to the requesting user. 8. The user request processing unit, together with the ranking information, the number of users who scored each evaluation target item, the average score of each evaluation target item, the variance of the score of each evaluation target item, and the evaluation of each evaluation target item. 8. The rating system according to claim 7, wherein related information including comments and recommendation information given by said user is also presented. 9. The requesting user's viewpoint is specified by a viewpoint vector indicating an evaluation target item regarded as important by the requesting user, and the requesting user displays a display object representing the evaluation target item on an origin on a two-dimensional plane. A relationship visualization unit that automatically extracts the viewpoint of the requesting user by allowing the viewpoint to be arranged around the object, and automatically converts the distance between the origin and each evaluation target item into a component of the viewpoint vector. The rating system according to claim 1, further comprising: 10. A similarity calculated by the evaluation history analysis unit is converted into a distance, and a display object representing the plurality of users is arranged on a two-dimensional plane while maintaining a distance relationship between the plurality of users. 2. The rating system according to claim 1, further comprising a relationship visualization unit that displays a similarity relationship between the plurality of users in the form of a similarity network obtained by the above. 11. (a) Upon receiving a request from a user, presenting evaluation information of a plurality of evaluation target items by a plurality of users to the requesting user, the requesting user can select an appropriate item from the plurality of evaluation target items. (B) analyzing the scoring histories of the plurality of users with respect to the evaluation target item based on a viewpoint specified by the requesting user, thereby analyzing the similarity between the plurality of users. And recommending an item deemed appropriate for the requesting user based on the similarity. 12. The viewpoint of the requesting user is specified by a viewpoint vector indicating an evaluation target item considered to be important by the requesting user. In the step (b), each user performs an evaluation on each evaluation target item. 12. The collaborative information filtering method according to claim 11, wherein the similarity defined by the given scoring value and the viewpoint vector is calculated. 13. In the step (b), a difference between a score value given to each evaluation target item by each user and an average value of all score values given to each evaluation target item by all users. The method according to claim 11, wherein the similarity defined by the following is calculated. 14. The collaborative information filtering method according to claim 11, wherein in step (b), a similar user of the requesting user is automatically determined based on the similarity. 15. The collaborative information filtering method according to claim 14, wherein in the step (b), an item having a high score by the user determined as the similar user is recommended. 16. In the step (b), the evaluation of an evaluation target item that has not been scored by each user is predicted based on the similarity and the evaluation of the evaluation target item by another user. 16. The method according to claim 15, wherein recommendation is made. 17. In the step (b), a ranking of the evaluation target item is also determined based on a score given to the evaluation target item by a plurality of users, and in the step (a), the ranking of the evaluation target item is determined. 12. The collaborative information filtering method according to claim 11, wherein ranking information on the ranking determined in b) is presented to the requesting user. 18. In the step (a), together with the ranking information, the number of users who scored each evaluation target item,
18. The method according to claim 17, further comprising presenting an average score of each evaluation target item, a variance of scores of each evaluation target item, and related information including comments and recommendation information given by a user who has scored each evaluation target item. Cooperative information filtering method. 19. The requesting user's viewpoint is specified by a viewpoint vector indicating an evaluation target item regarded as important by the requesting user, and the requesting user specifies a display object of the evaluation target item at an origin on a two-dimensional plane. Automatically extracting a viewpoint of the requesting user by enabling the surroundings to be arranged, and automatically converting a distance between the origin and each evaluation target item into a component of the viewpoint vector. Claim 11
Described collaborative information filtering method. 20. Converting the similarity calculated in the step (b) into a distance, and arranging display objects of the plurality of users on a two-dimensional plane while maintaining a distance relationship between the plurality of users. 12. The collaborative information filtering method according to claim 11, further comprising a step of displaying a similarity relationship between the plurality of users in a form of a similarity network.